The Impact of Regulation on Innovation
GAICIL Insights
For a French firm, crossing 50 employees brings a sharp increase in the burden of labor regulation.1 This paper shows the hidden price is paid in innovation never undertaken. The Impact of Regulation on Innovation, by Philippe Aghion, Antonin Bergeaud and John Van Reenen, builds a framework for the equilibrium impact of regulation on endogenous innovation with heterogeneous firms. They implement it with French firm-level panel data, exploiting that sharp increase in the burden of labor regulations on companies with 50 or more employees.
The evidence is compelling. Consistent with the model, the fraction of innovating firms falls just below the regulatory threshold. Firms just under the threshold also show a weaker innovation response to demand shocks. On average, firms innovate more after positive market size shocks from export markets, yet this relationship significantly weakens near the threshold. The prospect of regulatory costs discourages growth, and with it innovation. Firms near the line choose to stay small, and small firms plan small.
The numbers are significant. Regulation reduces aggregate innovation by 5.7 percent in the abstract's summary, and baseline estimates imply a lower bound on the welfare loss of about 2.3%. These losses exceed those of analyses treating technology as exogenous. When regulation affects innovation, its costs compound over time through slower growth. A compliance cost measured once is a compliance cost underestimated.
A nuance softens the blow slightly. Using citation information, the authors find that the labor regulation deters incremental innovation and has little effect on more radical innovation. That fits a generalization of the model allowing simultaneous investment in two types of R&D, and slightly mitigates the welfare loss. Firms near the threshold may shift toward more radical projects, or radical innovators may be less constrained by growth-related costs.
AI regulation is full of thresholds. Many AI rules impose obligations on firms or systems above certain scales. The paper shows size-dependent regulations can discourage firms from growing and innovating near the threshold. AI rules imposing significant costs at particular sizes may create similar distortions, reducing innovation among firms approaching the line. A compute or revenue cutoff may create a crowd of companies parked just beneath it.
The framework doubles as an evaluation kit. Examining innovation behavior around thresholds and firms' responses to demand shocks lets regulators estimate the innovation cost of size-dependent rules. Applying it to AI regulations as they take effect would generate evidence for refining them. Every threshold is a natural experiment waiting to be read.
The radical-innovation result offers both reassurance and a caution. If regulations mainly deter incremental innovation, their costs may be partly offset. Yet incremental innovation contributes substantially to productivity growth, and its loss remains costly. Regulators should aim to minimize distortions to both types. The steady improvements that make technology usable are easy to lose without anyone noticing.
The principle is to count dynamic costs. Static analyses measuring only immediate compliance costs understate the true burden when regulation affects innovation and growth. The authors' framework incorporates these dynamic effects and reveals larger welfare losses. For AI, where innovation is the central source of value, dynamic analysis is essential.
Design can reduce the damage. Phasing in obligations gradually as firms grow, avoiding sharp thresholds, limits the incentive to stay small. Targeting obligations at specific risks achieves regulatory goals with fewer distortions. These lessons apply directly to AI rules. A ramp is kinder to growth than a cliff.
Taken together, regulation can have substantial unintended effects on innovation. French labor regulations at the 50-employee threshold discouraged firms from innovating, with meaningful aggregate costs. As governments design rules for AI, they should take seriously the evidence that size-dependent regulations can deter growth and innovation. This paper provides a framework for measuring such effects and a reminder that the dynamic costs of regulation deserve careful attention.
For AI lawmakers drafting size cutoffs, the sting is this: plot the number of firms just below each threshold a few years after the rule takes effect. If a bulge appears, the rule is taxing growth itself, and its true cost is the innovation that bulge never produced.
Introduction
We present a framework that can be used to assess the equilibrium impact of regulation on endogenous innovation with heterogeneous firms. We implement this model using French firm-level panel data, where there is a sharp increase in the burden of labor regulations on companies with 50 or more employees. Consistent with the model’s qualitative predictions, we find a fall in the fraction of innovating firms just to the left of the regulatory threshold. Furthermore, we find a reduction in the innovation response of firms to demand shocks just below the threshold. Regulation reduces aggregate innovation by 5.7 percent. (JEL D22, K31, L11, L25, L51, O31, O34)
Takeaway
In this paper, we have developed a framework to analyze the impact of regulation on innovation. We applied this to France, where strong labor regulations affect firms who employ 50 or more workers. We showed both theoretically and empirically that the prospect of these regulatory costs discourages firms just below the threshold from innovating, where innovation is measured by the volume of patent applications. This relationship emerges both when looking non-parametrically at patent density around the threshold and in a parametric exercise where we examine the heterogeneous response of firms to exogenous market size shocks (from export markets). On average, firms innovate more when they experience a positive shock, but this relationship significantly weakens when a firm is just below the regulatory threshold. We then use moments from our data and the literature to calibrate the structural parameters in the model. For example, using estimates of the R&D cost function, we can back out the magnitude of the regulatory tax from the ratio between the slopes of the innovation-size relationship for large firms compared to small firms. Our baseline estimates imply an aggregate innovation (and therefore growth) loss of about 5.8% and a lower bound on the loss of welfare of about 2.3%. This suggests larger welfare losses than existing analyses that take technology as exogenous. A caveat to this conclusion is that when we use information on citations we find that the labor regulation deters incremental innovation, but has little effect on more radical innovation.1
This is consistent with a generalization of the model which allows for simultaneous investment in two types of R&D, and slightly mitigates the welfare loss of the regulation. The analysis in this paper can be extended in several directions. First, our focus in this paper was on the long-run steady state, but it is perhaps equally important to analyze the transitional dynamics triggered by policy changes, and to factor in adjustment costs. Second, the framework can be applied to many other countries and regulatory settings. Third, our analysis remained focused on the costs of the labor regulation. However, such a regulation may also bring benefits in the form of better insurance and deeper involvement of employees in the management of the firm, which in turn fosters trust between employers and employees. Future work should take such benefits into account to see if they are sufficient to overcome the costs we have identified here.
Notes
- Philippe Aghion, Antonin Bergeaud & John Van Reenen, The Impact of Regulation on Innovation, 113 Am. Econ. Rev. 2894 (2023), https://doi.org/10.1257/aer.20210107. Takeaway source: Conclusion, reproduced from a free author or repository copy (working-paper version). ↩
Cite as: Philippe Aghion, Antonin Bergeaud & John Van Reenen, The Impact of Regulation on Innovation, 113 Am. Econ. Rev. 2894 (2023), https://doi.org/10.1257/aer.20210107
Header image: Capitol Building 4 (27680176211).jpg by Tony Hisgett from Birmingham, UK, CC BY 2.0, via Wikimedia Commons.
.jpg){kind=link}